Google and Top SEO and GEO Studies agree: Schema is not a ranking factor

Schema Won’t Save You: What the Studies Actually Say

Adding schema markup does not get you ranked, and it does not get you cited by AI. Two controlled experiments published in 2026 tested it properly, and both came back flat.

For two years the pitch has been the same: add JSON-LD, help the machines “understand” you, collect your AI citations. Agencies sold schema audits as AEO and GEO packages. The evidence for it was a correlation and a lot of blog posts quoting each other.

Here is every study worth knowing about, including the ones that cut the other way.

Ahrefs: 1,885 pages, no lift

Ahrefs ran the first large controlled test of schema and AI citations, and adding schema produced no meaningful increase on any platform. Louise Linehan and Xibeijia Guan published it in May 2026.

They tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and matched them against about 4,000 control pages that never added it. They ran the analysis four different ways.

Platform	Change in citations after adding schema
Google AI Overviews	-4.6%
Google AI Mode	+2.4%
ChatGPT	+2.2%

The AI Overviews drop is statistically significant, but treated and control pages were both already declining, so Ahrefs does not pin it on schema. The other two are noise.

The same report explains where the myth came from. Across 6 million URLs, pages cited by AI were almost three times more likely to carry JSON-LD, and about 53% of cited pages ran schema. That is the stat on every conference slide. It is a correlation: sites that bother with schema also publish better content and earn more links.

One honest limit. Every page in the test already had 100+ AI Overview citations before schema went on. The study shows schema will not push a visible page higher. It does not show what schema does for a page AI has never noticed.

Jake Hundley: schema did nothing for local rankings

Jake Hundley of Evergrow Marketing ran a controlled experiment on LocalBusiness schema and found no ranking improvement on any traditional search engine. The details below come from CMSWire’s write-up of the study.

The design is what makes it worth citing:

29 client domains across 36 locations, all landscaping and lawn care, so industry is not a variable
Split into a control group and a test group
All schema stripped from every site, then a five-week reset
Advanced LocalBusiness schema added to the test group’s homepages only, then five more weeks of tracking
Methodology peer-reviewed before the results were known, by schema advocates and skeptics alike
A one-tailed Welch’s t-test, with 90% confidence as the bar for a real result

Google, Google Mobile, Bing and Yahoo rankings did not move. Google Maps rankings did not move. Share of Local Voice did not move.

The one exception was ChatGPT. Positioning improved by roughly 3.3 positions at 92.91% confidence, and Share of AI Voice rose about 10 percentage points at 91.51% confidence. Nobody has established why. So the accurate summary is: zero for search, a possible edge in ChatGPT for local businesses, in one industry, from one study.

Google has said it repeatedly

Google’s own documentation says structured data is not required for AI search and is not a ranking factor.

The generative AI optimization guide. Published in May 2026, it has a mythbusting section listing what publishers do not need to do. Structured data is on the list: not required for generative AI search, and no special schema.org markup to add. Google also says to keep using it, because it still makes pages eligible for rich results. Coverage here.
John Mueller. He said on Bluesky that structured data will not make a site rank better and is only used to display supported search features. Search Engine Roundtable has the post.
The deprecations. Google keeps retiring structured data types, including FAQ rich results for most sites. A platform that depended on your markup would not keep removing support for it.

None of this is new. People just prefer the blog posts.

SE Ranking and Semrush: read these carefully

Neither of these is a clean debunk, and anyone who tells you otherwise has not read them.

SE Ranking analyzed 129,000 domains and 216,524 pages across 20 niches to find what correlates with ChatGPT citations. FAQ schema underperformed: pages with it averaged 3.6 citations. Links, traffic and rankings were what tracked with citations. This is correlational and covers FAQ schema only. Search Engine Journal covered it.

Semrush found the opposite direction. Its content optimization study, published January 2026, reported a +21.6% positive correlation between structured data elements and AI citations.

That Semrush number is the same kind of evidence as the Ahrefs 6 million URL finding: a correlation. Ahrefs then tested whether adding schema causes the citations, and it did not. A correlation study cannot overrule a controlled one.

Why it doesn’t work: the models read the page, not the markup

When AI tools fetch a page live, the tests so far show they read the visible text and ignore the JSON-LD.

Test	What it found
searchVIU, October 2025	Tested ChatGPT, Claude, Perplexity, Gemini and Google AI Mode. On direct fetch, none used JSON-LD; they extracted visible HTML only. Summary
Dejan	Two identical pages, one with schema and one without. No evidence the schema data ever reached the model through ChatGPT’s browsing tool.
Mark Williams-Cook, September 2025	Challenged the claim that an AI returning a schema-only fact proves it is using schema. The models picked up the text, not the structure.
Search Atlas	Compared schema coverage with domain visibility in OpenAI, Gemini and Perplexity. No measurable effect at any level of schema adoption.

searchVIU notes the limit: schema could still play a role at indexing time, or inside search-integrated systems like AI Overviews and Bing Copilot. These tests cover live retrieval, not every pipeline.

The other side, and why it doesn’t settle anything

There is pro-schema evidence. None of it comes from a controlled test at scale.

The Reddit “Schema vs No-Schema” experiment. A sandbox test posted to Reddit concluded that Google’s AI trusts machine-readable data over on-page text for factual queries. It is one site with no control group.
AirOps and Kevin Indig. Across 16,851 queries, pages with JSON-LD were cited by ChatGPT 38.5% of the time versus 32.0% without. That is a comparison of pages with and without schema, not a before-and-after. Coverage.
Growth Marshal. Attribute-rich schema earned a 61.7% citation rate, but generic schema did worse than none: 41.6% versus 59.8%. If anything that argues against the plugin-default markup most sites run. Coverage.
Hundley’s ChatGPT result. The local ChatGPT lift above is real data and deserves a replication.

The open question is whether schema helps a page that is not yet cited. Nobody has run that test with a control group. Until someone does, “it might help undiscovered pages” is a hypothesis.

What to do with this

Keep schema where it produces something you can see in the results, and stop selling or buying it as a ranking or AI visibility lever.

Keep it for Product, Review, Recipe, Event and JobPosting markup. Those generate rich results, and rich results earn clicks.
Add the basics like Organization and LocalBusiness. They take minutes and do no harm. If you are a local business that cares about ChatGPT, Hundley’s data gives you a reason.
Stop paying for schema audits pitched as AEO or GEO. The one controlled test at scale found no citation lift.
Test it yourself. Take pages with low or no AI citations, add schema to half, leave the other half alone, and compare after 30 and 60 days. Without the control group you are measuring the weather.

AI Overviews and AI Mode retrieve from Google’s index. ChatGPT and the rest read the text on your page. In both cases what gets you cited is ranking, authority and content that answers the query, which is what got you cited before anyone renamed SEO.

Weblinkr